[llvm] 14c3ab9 - IRBuilder: Fix not handling strictfp minnum/maxnum

[Matt Arsenault via llvm-commits]

Author: Matt Arsenault
Date: 2023-07-11T18:51:50-04:00
New Revision: 14c3ab945be9c49964dbf79f13d8ff8df1ff7b72

URL: https://github.com/llvm/llvm-project/commit/14c3ab945be9c49964dbf79f13d8ff8df1ff7b72
DIFF: https://github.com/llvm/llvm-project/commit/14c3ab945be9c49964dbf79f13d8ff8df1ff7b72.diff

LOG: IRBuilder: Fix not handling strictfp minnum/maxnum

Removing the rounding mode arguments seems like more trouble than
it's worth. minimum and maximum are still broken.

https://reviews.llvm.org/D154994

Added: 
    

Modified: 
    llvm/include/llvm/IR/IRBuilder.h
    llvm/lib/IR/IRBuilder.cpp
    llvm/test/Transforms/AtomicExpand/AMDGPU/expand-atomic-rmw-fmax.ll
    llvm/test/Transforms/AtomicExpand/AMDGPU/expand-atomic-rmw-fmin.ll

Removed: 
    


################################################################################
diff  --git a/llvm/include/llvm/IR/IRBuilder.h b/llvm/include/llvm/IR/IRBuilder.h
index ebce038201d958..28380a3f606f2c 100644
--- a/llvm/include/llvm/IR/IRBuilder.h
+++ b/llvm/include/llvm/IR/IRBuilder.h
@@ -956,11 +956,21 @@ class IRBuilderBase {
 
   /// Create call to the minnum intrinsic.
   CallInst *CreateMinNum(Value *LHS, Value *RHS, const Twine &Name = "") {
+    if (IsFPConstrained) {
+      return CreateConstrainedFPUnroundedBinOp(
+          Intrinsic::experimental_constrained_minnum, LHS, RHS, nullptr, Name);
+    }
+
     return CreateBinaryIntrinsic(Intrinsic::minnum, LHS, RHS, nullptr, Name);
   }
 
   /// Create call to the maxnum intrinsic.
   CallInst *CreateMaxNum(Value *LHS, Value *RHS, const Twine &Name = "") {
+    if (IsFPConstrained) {
+      return CreateConstrainedFPUnroundedBinOp(
+          Intrinsic::experimental_constrained_maxnum, LHS, RHS, nullptr, Name);
+    }
+
     return CreateBinaryIntrinsic(Intrinsic::maxnum, LHS, RHS, nullptr, Name);
   }
 
@@ -1658,6 +1668,11 @@ class IRBuilderBase {
       std::optional<RoundingMode> Rounding = std::nullopt,
       std::optional<fp::ExceptionBehavior> Except = std::nullopt);
 
+  CallInst *CreateConstrainedFPUnroundedBinOp(
+      Intrinsic::ID ID, Value *L, Value *R, Instruction *FMFSource = nullptr,
+      const Twine &Name = "", MDNode *FPMathTag = nullptr,
+      std::optional<fp::ExceptionBehavior> Except = std::nullopt);
+
   Value *CreateNeg(Value *V, const Twine &Name = "", bool HasNUW = false,
                    bool HasNSW = false) {
     return CreateSub(Constant::getNullValue(V->getType()), V, Name, HasNUW,

diff  --git a/llvm/lib/IR/IRBuilder.cpp b/llvm/lib/IR/IRBuilder.cpp
index 491065dea29ffb..a015073f9b3676 100644
--- a/llvm/lib/IR/IRBuilder.cpp
+++ b/llvm/lib/IR/IRBuilder.cpp
@@ -1032,6 +1032,23 @@ CallInst *IRBuilderBase::CreateConstrainedFPBinOp(
   return C;
 }
 
+CallInst *IRBuilderBase::CreateConstrainedFPUnroundedBinOp(
+    Intrinsic::ID ID, Value *L, Value *R, Instruction *FMFSource,
+    const Twine &Name, MDNode *FPMathTag,
+    std::optional<fp::ExceptionBehavior> Except) {
+  Value *ExceptV = getConstrainedFPExcept(Except);
+
+  FastMathFlags UseFMF = FMF;
+  if (FMFSource)
+    UseFMF = FMFSource->getFastMathFlags();
+
+  CallInst *C =
+      CreateIntrinsic(ID, {L->getType()}, {L, R, ExceptV}, nullptr, Name);
+  setConstrainedFPCallAttr(C);
+  setFPAttrs(C, FPMathTag, UseFMF);
+  return C;
+}
+
 Value *IRBuilderBase::CreateNAryOp(unsigned Opc, ArrayRef<Value *> Ops,
                                    const Twine &Name, MDNode *FPMathTag) {
   if (Instruction::isBinaryOp(Opc)) {

diff  --git a/llvm/test/Transforms/AtomicExpand/AMDGPU/expand-atomic-rmw-fmax.ll b/llvm/test/Transforms/AtomicExpand/AMDGPU/expand-atomic-rmw-fmax.ll
index 12441d3009bc10..5c4462caafb84e 100644
--- a/llvm/test/Transforms/AtomicExpand/AMDGPU/expand-atomic-rmw-fmax.ll
+++ b/llvm/test/Transforms/AtomicExpand/AMDGPU/expand-atomic-rmw-fmax.ll
@@ -268,7 +268,7 @@ define double @test_atomicrmw_fmax_f64_global_strictfp(ptr addrspace(1) %ptr, do
 ; GCN-NEXT:    br label [[ATOMICRMW_START:%.*]]
 ; GCN:       atomicrmw.start:
 ; GCN-NEXT:    [[LOADED:%.*]] = phi double [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP6:%.*]], [[ATOMICRMW_START]] ]
-; GCN-NEXT:    [[TMP2:%.*]] = call double @llvm.maxnum.f64(double [[LOADED]], double [[VALUE:%.*]]) #[[ATTR3:[0-9]+]]
+; GCN-NEXT:    [[TMP2:%.*]] = call double @llvm.experimental.constrained.maxnum.f64(double [[LOADED]], double [[VALUE:%.*]], metadata !"fpexcept.strict") #[[ATTR4:[0-9]+]]
 ; GCN-NEXT:    [[TMP3:%.*]] = bitcast double [[TMP2]] to i64
 ; GCN-NEXT:    [[TMP4:%.*]] = bitcast double [[LOADED]] to i64
 ; GCN-NEXT:    [[TMP5:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i64 [[TMP4]], i64 [[TMP3]] seq_cst seq_cst, align 8

diff  --git a/llvm/test/Transforms/AtomicExpand/AMDGPU/expand-atomic-rmw-fmin.ll b/llvm/test/Transforms/AtomicExpand/AMDGPU/expand-atomic-rmw-fmin.ll
index 3eda559f83fdd8..0e32489286a16a 100644
--- a/llvm/test/Transforms/AtomicExpand/AMDGPU/expand-atomic-rmw-fmin.ll
+++ b/llvm/test/Transforms/AtomicExpand/AMDGPU/expand-atomic-rmw-fmin.ll
@@ -268,7 +268,7 @@ define double @test_atomicrmw_fmin_f64_global_strictfp(ptr addrspace(1) %ptr, do
 ; GCN-NEXT:    br label [[ATOMICRMW_START:%.*]]
 ; GCN:       atomicrmw.start:
 ; GCN-NEXT:    [[LOADED:%.*]] = phi double [ [[TMP1]], [[TMP0:%.*]] ], [ [[TMP6:%.*]], [[ATOMICRMW_START]] ]
-; GCN-NEXT:    [[TMP2:%.*]] = call double @llvm.minnum.f64(double [[LOADED]], double [[VALUE:%.*]])
+; GCN-NEXT:    [[TMP2:%.*]] = call double @llvm.experimental.constrained.minnum.f64(double [[LOADED]], double [[VALUE:%.*]], metadata !"fpexcept.strict") #[[ATTR4:[0-9]+]]
 ; GCN-NEXT:    [[TMP3:%.*]] = bitcast double [[TMP2]] to i64
 ; GCN-NEXT:    [[TMP4:%.*]] = bitcast double [[LOADED]] to i64
 ; GCN-NEXT:    [[TMP5:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i64 [[TMP4]], i64 [[TMP3]] seq_cst seq_cst, align 8